A23: Evaluation of Data-Intensive Applications on Intel Knights Landing Cluster
Student: Tao Gao (University of Delaware)
Supervisor: Michela Taufer (University of Delaware)
Abstract: Analyzing and understanding large datasets on high performance computing platforms is becoming more and more important in various scientific domains. MapReduce is the dominant programming model for processing these datasets. Platforms for data processing are empowered by many-core nodes with cutting-edge processing units. Intel Knights Landing (KNL) is the new arrival in the field. However, this new architecture has not been fully evaluated for data-intensive applications. In this poster, we present the assess of KNL on the performance of three key data-intensive applications based on a high-performance MapReduce programming framework on the latest KNL-cluster, Stampede2. We focus on the impact of different KNL memory models, we compare Stampede2 with other clusters such as Tianhe-2 and Mira, and we measure the scalability of large datasets. We observe how KNL-based clusters are a promising architecture for data-intensive applications. We also identify key aspects to enable more efficient usage of KNL-based clusters.
ACM-SRC Semi-Finalist: no
Poster: pdf
Two-page extended abstract: pdf
Poster Index